In recent years, in the field of cluster system which combines multiple computers into a group, diskless cluster is attracting attention wherein individual computers constituting the cluster system contain no disk. In such a diskless cluster, root file systems of each computer are managed by an image server. These root file systems are mounted in each computer, whereby the computers are operated.
Diskless operation facilitates the management of each computer. However, when the image server has too much traffic, performance of the system degrades.
Therefore, for systems where performance has priority, there has been proposed an operation using a local hard disk drive (e.g., refer to Patent Document 1, Published Unexamined Japanese Patent Application No. 9-034825). In a client-server configuration as being illustrated in FIG. 3 of Patent Document 1, each client includes a local root file memory which stores a copy of local root file system from the root file memory connected to a server.
Also for a diskless computer, there are: a technique (e.g., refer to Patent Document 2, Published Unexamined Japanese Patent Application No. 8-328912) by which stack and register contents are sent and saved as a core image to a remote server when a failure occurs in the application; and a technique by which a local non-volatile memory cache is used when the network cannot be used temporarily or is low in data transmission speed, and the memory cache is synchronized with the storage volume of the server after the status of network has been improved (e.g., refer to Patent Document 3, Published Unexamined Japanese Patent Application No. 2004-178596).
However, in the operation of using local hard disk drive as being illustrated in Patent Document 1, failures may occur in the local hard disk drive. When a failure occurs, a new hard disk drive must be prepared and used to restore, or reconstruct, the system environment, which requires time and labor.
In addition, when a failure of the hard disk drive occurs during execution of an application, it is difficult to change back the application state to the pre-failure state.